Усилитель ULTIMATE

[MAXIM_A]

Господа-товарищи соорудили классный усилок, слабо попробывать повторить?
Я хочу, начну с оконечника.
The project scope was defined in the early 2007.
Key performance requirements include:
• Class AB topology, 100W into 8 ohm and 200W into 4 ohm
• Bandwidth of minimum 20Hz to 20KHz +/-0.1dB
• Under 1 ppm (80KHz bandwidth) THD (that is, 0.0001%)
at all frequencies between 20Hz and 20KHz and output
powers between 1W and 200W
• Under 1 ppm 19+20KHz CCIF IMD (that is, 0.0001%) at all
output powers between 1W and 200W
http://www.synaesthesia.ca/index.html

Печатная плата ВК:

[Prohoji]

sia_2, Я сегодня на дне рождения Но не могу удержаться чтобы не сказать за сверх глубокую ось. Я её лично СЛЫШАЛ. Мож я и ошибаюсь (ну эт я так .. для понта...) Но помому это абсолютный выход. Короче время покажет . Кажись NIK сделал всех..

[RED FOX]

Всё-таки Nick.

[syn08]

LM318 has only 12nV/Hz^1/2 voltage noise density up to 100 kHz and less than 0.3 pA/Hz^1/2 current noise density. Flicker voltage noise corner less than 300Hz. This opamp has significant HF noise, but it isn*t critical for this application (EIN < 1.5 uVA). SNR value of whole amp 114...116 dBA, depending from gain.

About 12nV/rtHz is the plateau noise at Rs=100ohm At 100kohm it has 60nV/rtHz.

But let*s consider 12nV/rtHz. The rule of thumb is that a 5 nV/rt Hz input noise in an unweighted 20 khz bandwidth yields an S/N with respect to 1 V rms input of 120 dB. The LM118 12nV/rtHz would entirely cover the difference to your measured S/N, but you also have to consider two more noise sources: the source impedance and the feedback network impedance. A 1kohm feedback resistor will add another 4nV/rtHz to the noise budget. Only these two components will take the S/N ratio in your amp down to under 110dB and again, this is before considering any other noise sources. I can*t explain the difference to your measurements, could it be the old AP1 resolution?

In the PGP amp, by using 2SA970/2SC2240 ultra low noise devices, the feedback network impedance is clearly dominating the noise. I have measured anywhere between 116dB and 120dB but I don*t really trust these mesurements, as they are at the limit of my equipment; they seem to be a little optimistic. But then again, I think we are discussing here about angel*s sex, 110dB or 120dB S/N, who cares? The bottom line is that any S/N claim over 120dB is bullshit.

Regarding the overcurrent protection, I have tried something very similar with what you have in your amp quite some time ago and the results were dissapointing. The protection network was either impacting the sound, or it was unable to protect against output shorts. The only explanation that I may have is that your amp seems to be very conservative regarding the SOA of the output devices. Certainly, sharing the current among so many output devices helps in keeping very large SOA margins, in which the protection may trigger before affecting (at normal output levels) the sound. Which again boils down to building an amp that is able to drive 1ohm continuously, to provide a very large operation margin. This seems to be a viable engineering approach, although from an economics perspective it looks quite expensive.

[sia_2]

sia_2, Я сегодня на дне рождения Но не могу удержаться чтобы не сказать за сверх глубокую ось. Я её лично СЛЫШАЛ. Мож я и ошибаюсь (ну эт я так .. для понта...) Но помому это абсолютный выход. Короче время покажет . Кажись NIK сделал всех..

Ну так к такому выводу пришли в Белл Лабс более 70 лет назад. Мы всего "лишь делаем, как написано до нас"

The rule of thumb is that a 5 nV/rt Hz input noise in an unweighted 20 khz bandwidth yields an S/N with respect to 1 V rms input of 120 dB.

This "rule of thumb" slightly pessimistical - 5 nV/rtHz corresponds to 0.71 uV RMS unweighted @ 20 kHz, 0r -123 dB re 1 V RMS. May be, you wrote it about 0 dBU (approx. 760 mV) reference level ? Then it could be acceptable.
Добавлено через 1 час 23 минуты

About 12nV/rtHz is the plateau noise at Rs=100ohm At 100kohm it has 60nV/rtHz.

These 60 nV/rtHz @100 K is good value and demonstrate low current noise - intrisinc thermal noise density of 100 K resistor is 40 nV/rtHz. Total noise - 60 nV/rtHz (Noise factor around 3 dB), and current noise can be estimated as SQRT(60^2-40^2-12^2)/100 ~ 0.4 pA/rtHz. Direct (and much more precision) measurements by mean of 5 M cooled resistors give results 0.25...0.28 pA/rtHz.
Virtually all noise of UL amp generated into resistors in input filter-preamplifier, because input impedance couldn*t lowered than 10...12 K due to overloading of previous stages. Therefore opamp with lower voltage noise has no sense for this application.

Simple calculations:

Filter noise gain is 3, input/feedback parallel impedance is 8 K (24 || 12), series to opamp input - 10 K. Total noise resistance - 18K. Total filter output noise density (at shorted input) is
3*SQRT((18*(4^2))+(18*0.28)^2+12^2) = 64.2 nV/rtHz (RTI 32.1 nV/rtHz). Contribution of Op-Amp noise less than one-third by power.

Main amp noise gain is 6.76, total noise resistance 3K (3.47K || 20K).
Output noise density at shorted main amp input is
6.76*SQRT((3*(4^2))+(3*0.28)^2+12^2) = 94 nv/rtHz (16.3 nV/rtHz RTI)

Whole amp RTI noise is SQRT(64.2^2+16.3^2)/2 = 33.1 nV/rtHz.
This value equal to 4.67 uV lin, and 3.3 uV "IEC A".
Overall gain of this amp around 11.5 (2*5.76), or 21.2 dB.
Nominal output voltage at 45V supply is 28V eff, and corresponding input voltage is 2.44V (28/11.5).

2.44V/4.67uV = 522400, or 114 dB Lin
2.44V/3.3uVA = 739000, or 117 dB "IEC A"

Really measured values simply corresponds with abovementioned calculations.

In the PGP amp, by using 2SA970/2SC2240 ultra low noise devices, the feedback network impedance is clearly dominating the noise. I have measured anywhere between 116dB and 120dB but I don*t really trust these mesurements, as they are at the limit of my equipment; they seem to be a little optimistic. But then again, I think we are discussing here about angel*s sex, 110dB or 120dB S/N, who cares? The bottom line is that any S/N claim over 120dB is bullshit.

If you can*t get good values, it does not mean that it is impossible. Many measurement equipment has S/N ratio much more 120 dB.
Halcro DM58 has measured EIN around 0.42 uV "IEC A". From other point of view, values better than 110 dB haven*t techical reasons, only marketing .

Regarding the overcurrent protection, I have tried something very similar with what you have in your amp quite some time ago and the results were dissapointing. The protection network was either impacting the sound, or it was unable to protect against output shorts.

You didn*t take the principle of this protection - if current sense transistors begin works, trigger circuit on VT9-10 latched and off VAS completely. But when trigger circuit isn*t latched, no any influence to cignal chain.
This circuit slightly slow, but output current can*t rise too fast due to output inductors.

This seems to be a viable engineering approach, although from an economics perspective it looks quite expensive.

Bipolar output devices are virtually cheapest items on BOM, BD911/912 costs only around 40-50c (!). Therefore its very rational choice.

[syn08]

C B0: : B0:><C 2K2>4C ?@8H;8 2 5;; 01A 1>;55 70 ;5B =0704. K 2A53> ";8HL 45;05<, :0: =0?8A0=> 4> =0A"

>102;5=> G5@57 1 G0A 23 <8=CBK

These 60 nV/rtHz @100 K is good value and demonstrate low current noise - intrisinc thermal noise density of 100 K resistor is 40 nV/rtHz. Total noise - 60 nV/rtHz (Noise factor around 3 dB), and current noise can be estimated as SQRT(60^2-40^2-12^2)/100 ~ 0.4 pA/rtHz. Direct (and much more precision) measurements by mean of 5 M cooled resistors give results 0.25...0.28 pA/rtHz.
Virtually all noise of UL amp generated into resistors in input filter-preamplifier, because input impedance couldn*t lowered than 10...12 K due to overloading of previous stages. Therefore opamp with lower voltage noise has no sense for this application.

Simple calculations:

Filter noise gain is 3, input/feedback parallel impedance is 8 K (24 || 12), series to opamp input - 10 K. Total noise resistance - 18K. Total filter output noise density (at shorted input) is
3*SQRT((18*(4^2))+(18*0.28)^2+12^2) = 64.2 nV/rtHz (RTI 32.1 nV/rtHz). Contribution of Op-Amp noise less than one-third by power.

Main amp noise gain is 6.76, total noise resistance 3K (3.47K || 20K).
Output noise density at shorted main amp input is
6.76*SQRT((3*(4^2))+(3*0.28)^2+12^2) = 94 nv/rtHz (16.3 nV/rtHz RTI)

Whole amp RTI noise is SQRT(64.2^2+16.3^2)/2 = 33.1 nV/rtHz.
This value equal to 4.67 uV lin, and 3.3 uV "IEC A".
Overall gain of this amp around 11.5 (2*5.76), or 21.2 dB.
Nominal output voltage at 45V supply is 28V eff, and corresponding input voltage is 2.44V (28/11.5).

2.44V/4.67uV = 522400, or 114 dB Lin
2.44V/3.3uVA = 739000, or 117 dB "IEC A"

Really measured values simply corresponds with abovementioned calculations.

If you can*t give good values, itsn*t impossibleness. Many measurement equipment has S/N ratio much more 120 dB.
Halcro DM58 has measured EIN around 0.42 uV "IEC A". In other point of view, values better than 110 dB haven*t techical reasons, only marketing .

You didn*t take the principle of this protection - if current sense transistors begin works, trigger circuit on VT9-10 latched and off VAS completely. But when trigger circuit isn*t latched, no any influence to cignal chain.
This circuit slightly slow, but output current can*t rise too fast due to output inductors.

Bipolar output devices are virtually cheapest items on BOM, BD911/912 costs only around 40-50c (!). Therefore its very rational choice.

You are using a different definition for SNR. Mine do not include the amp gain, that is, its normalized to the gain. Therefore you are not comparing apples to apples.

I understand perfectly how your protection works. "no influence on the signal chain" that*s not true according to my measurements but then I do not have a better solution and its obviously better than nothing. The impact is due to the nonlinear capacitances which impacts, close to the tripping point, the DIM and TIM.

About costs, we live on different sides of the pond. All I can get here is 2SC5200/2SA1943 for $5 a pair plus taxes. Quite a difference...

[Nick]

The impact is due to the nonlinear capacitances which impacts, close to the tripping point, the DIM and TIM.

I think it is not true for this amp. The C41 and C42 capacitors virtually elemenates all problems with nonlinear capacitance coupling, especially if you take into account that distortion of VAS plus the output stage is more than 0.05%...

[sia_2]

You are using a different definition for SNR. Mine do not include the amp gain, that is, its normalized to the gain. Therefore you are not comparing apples to apples.

I understand perfectly how your protection works. "no influence on the signal chain" that*s not true according to my measurements but then I do not have a better solution and its obviously better than nothing. The impact is due to the nonlinear capacitances which impacts, close to the tripping point, the DIM and TIM.

About costs, we live on different sides of the pond. All I can get here is 2SC5200/2SA1943 for $5 a pair plus taxes. Quite a difference...

BD911 prices from catalog sellers

http://www.mouser.com/Search/Refine....user_Wildcards
http://search.digikey.com/scripts/Dk...?name=BD911-ND

2SC5200 prices from catalog sellers

http://search.digikey.com/scripts/Dk...5200&x=28&y=18
http://www.mouser.com/Search/Refine....user_Wildcards

SNR standard definition

SNR = 20 log ((max.RMS undistorted output voltage)/(RMS output noise into BW of interest))
or
SNR = 20 log ((max.RMS undistorted input voltage)/(RMS RTI noise into BW of interest)).
Both are eqivalent.
See also
http://en.wikipedia.org/wiki/Signal_to_noise

[syn08]

BD911 prices from catalog sellers

http://www.mouser.com/Search/Refine....user_Wildcards
http://search.digikey.com/scripts/Dk...?name=BD911-ND

2SC5200 prices from catalog sellers

http://search.digikey.com/scripts/Dk...5200&x=28&y=18
http://www.mouser.com/Search/Refine....user_Wildcards

SNR standard definition

SNR = 20 log ((max.RMS undistorted output voltage)/(RMS output noise into BW of interest))
or
SNR = 20 log ((max.RMS undistorted input voltage)/(RMS RTI noise into BW of interest)).
Both are eqivalent.
See also
http://en.wikipedia.org/wiki/Signal_to_noise

Well, I give up I do not plan to buy 2000pcs of BD911 to get a decent price and besides, whatever you are going to tell me about, building amps with 3MHz Ft output devices is not something that I would enjoy. I have my doubts about building anything very high performance and stable with such 3MHz output devices, but then again, what do I know? With such output devices, I find your loop gain at 25 KHz of about 90 dB (that is, about 750MHz GBW!) a truly amazing result. My amp has a measly 700KHz open loop bandwidth and a GBW of only 3.5MHz.

I*ll let you do the math about how much is 5nV/rtHz (and then 34nV/rtHz) into 20KHz unweighted, with respect to 1V, which is what I originally stated. May I suggest building an amp with an overall gain of 1? Following your calculations, that amp will certainly have a much better SNR.

I still believe that your protection may work well simply because the very large margin of SOA that you allowed.

Good luck guys and keep up the good work!

[Nick]

May I suggest building an amp with an overall gain of 1? Following your calculations, that amp will certainly have a much better SNR

Of course it will, and it should.

Добавлено через 2 минуты

I*ll let you do the math about how much is 5nV/rtHz (and then 34nV/rtHz) into 20KHz unweighted, with respect to 1V,

Why not with respect to 2 V or 5V ? I do not see any sence to measure or calculate SNR for power amp with respect to 1V...

Добавлено через 5 минут

I have my doubts about building anything very high performance and stable with such 3MHz output devices

It certainly possible, because in such schematics the signal from drivers can go thru the output stage (if it does not amplify). It is not needed to have power (current) amplification on 3 or 5 MHz for audio applications, so with appropriate output filter you can have enough phase margine at these frequencies even with such a slow transistors .

[Стрелец]

Господа-товарищи соорудили классный усилок, слабо попробывать повторить?
Я хочу, начну с оконечника.
The project scope was defined in the early 2007.
Key performance requirements include:
• Class AB topology, 100W into 8 ohm and 200W into 4 ohm
• Bandwidth of minimum 20Hz to 20KHz +/-0.1dB
• Under 1 ppm (80KHz bandwidth) THD (that is, 0.0001%)
at all frequencies between 20Hz and 20KHz and output
powers between 1W and 200W
• Under 1 ppm 19+20KHz CCIF IMD (that is, 0.0001%) at all
output powers between 1W and 200W
http://www.synaesthesia.ca/index.html

Печатная плата ВК:

MAXCIM_A, Вы собрали этот выходной каскад по этой плате, каковы результаты? Выложите схему , а то ссылка которую вы дали не работает.

[MAXIM_A]

MAXCIM_A, Вы собрали этот выходной каскад по этой плате, каковы результаты? Выложите схему , а то ссылка которую вы дали не работает.

Нет, спецы отговорили, а жаль.
Всё открывается, а схему попробую показать.

[kaskod]

а схему попробую показать

Зачем там столько повторителей ???

[MAXIM_A]

Зачем там столько повторителей ???

Почитайте теорию, если сможете, для линеризации ВК.

[kaskod]

Почитайте теорию, если сможете, для линеризации ВК.

А ну ну , а если не смогу ?

Offтопик:
Не грубите а слегка подумайте для какой нахрен лениризации Достаточно одного ЭП для 1058/162 В знании вами теории я нисколько "несомневаюсь"

Очередной огород И сильно неудивлюсь если вся эта лабуда загенерит, правда ребята понастарались напихать деталек шоб оно не генерило , но вот поможет ли ?

[renovator]

Очередной огород И сильно неудивлюсь если вся эта лабуда загенерит.

А тут, что пишут по этому поводу? http://www.synaesthesia.ca/OP-stage.html

В знании вами теории я нисколько "несомневаюсь"

По моему, эту конструкцию делали люди, с весьма сильной базовой подготовкой.

[kaskod]

По моему, эту конструкцию делали люди, с весьма сильной базовой подготовкой.

Сейчас помоделю и посмотрим имеет ли оно преимущества по сравнению с 1058/162 +одЫн ЭП , а если убрать все приблуды чтоб не генерили 1058/162.
В простейшем случае имеем 0,15 % и на 20кГц в том числе- по симмулятору без истоковых резисторов, что есть немного ниже чем в реалности-реально я получал около 0,25 % эти транзисторы ,1058/162, я крутил как только мог за что и полюбил. Второй вариант чуть позже если успею сегодня. Заметте это без ООС икажения самого истокового поовторителя на 1058/162 .Добавление тупо третьего Э.повторителя даёт увеличение Кг до 0,32 %, а вот теперь добавим "линеализатор-гармонизатор ". И не забыть меандр влупить тому усику , в ущерб чему там компенсация происходит? На скорую руку просмотрел -полное ФУФЛО.
---------------------------
Потрудитесь хотя бы помоделить весь этот бред, можно ещё поиграться ,но нах результат 0,45 % на 1КгЦ далее всю эту лабуду рассмативать стало неинтересно. Не надо валить на симмулятор, расхождения не более 10 % с данными моделями-проверено в железе. Меандр выглядит "получше" но это следствие что повторитель менее скоростной чем простой, посему нет выброса.
------------------
Короче: игра нестоит свеч, игра воображения не более